At this year’s International Solid State Circuits Conference in San Francisco, CA, MIT researchers unveiled the newest AI development – bringing neural networks to mobile devices via a low-power, local smart chip. It could be commercialized in the near future, vastly enhancing deep learning technology for good. (via MIT)
Innovators raise the bar. At the International Solid State Circuits Conference 2016, MIT researchers unveiled a custom chip named Eyeriss that could make neural networks (or Artificial Intelligence) a reality on mobile devices.
Neural networks are incredibly complex systems that mimic the functionality of the human brain. Mega complex computers are used to power these machines that push technology to further heights in the hopes of knowing what truly separates man and machine. With Eyeriss, these networks can be powered on low-power devices off-line, making for entirely new ways to interact with and study AI and deep learning.
MIT researchers worked hard to give Eyeriss comparable stats to its more power-hungry counterparts, and although the novel chip features 168 core graphics processors, it requires 10 times less energy than mobile graphics processors currently in production. It does this by prioritizing local communication, allowing function without access to internet or wasting energy.
Each of Eyeriss’ cores has its own memory and is capable of compressing data. Cores can also communicate with neighboring cores without requiring internet access, and will store new information locally to avoid spending energy. The chip does have a central core that acts as the main memory storage, but the chip itself is programmed to act locally when possible and it is also programmed with energy-saving processes to ensure the most energy efficient practices when possible.
The chip is revolutionary for all kinds of emerging technology. The Internet of Things is just one area of innovation that could significantly benefit from Eyeriss. Currently, security threats are a serious concern of IoT technology and devices – one that hasn’t truly been addressed. If IoT devices borrowed Eyeriss’ philosophy of acting locally, consumer information could be kept much more private by avoiding the constant transmission of data to the internet.
Another technology sector that could benefit from the Eyeriss technology is deep learning (i.e. machine learning). Deep learning could result in the development of a number of interactive mobile technologies, including facial, speech, and object recognition like never before. More than anything, we could see AI progress to custom-fit the unique needs of users everywhere.
The concept of artificial intelligence continues to surface in talk about the future of technology. Presently, it may provide a way for people to customize the way in which they engage with technology. If a phone could be customized to respond to the needs of the user specifically, each phone would become unique over time. Technology would provide the ultimate support for its users in daily life – something scientists have been working on since day one.
If we could develop relationships with our technology, we could provide an artificial system of support for even the most severe outcasts of our society. What would life be like if we could not only depend on one another, but on machines to ensure every person felt accepted, cared for, loved? It certainly isn’t a novel idea, but advancements like Eyeriss continue to push us to consider the very core of what makes us human.
The MIT research team that developed Eyeriss includes a number of accomplished researchers, including distinguished Assistant Professor of Electrical Engineering and Computer Science Vivienne Sze, Senior Distinguished Research Scientist at NVidia and MIT Professor in the Practice Joel Emer, postdoc researcher Tushar Krishna, and graduate student Yu-Hsin Chen. The project was partially funded by DARPA.
There is no word on when or if Eyeriss will be commercialized, but it is certainly a possibility. If nothing else, it opens doors to new technological possibilities of which we have only ever dreamed.
Have a story tip? Message me at: